Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.

Approach to get 128D face embeddings?

idata
Employee
1,299 Views

How should I proceed if I need 128D represention of faces just as in dlib and openface?

 

What needs to done/adapted/changed ? Please outline the steps. It will help in understanding the Movidius NCS better
0 Kudos
4 Replies
idata
Employee
1,012 Views

@gauravmoon Assuming that you are referring to a network like FaceNet, the current NCSDK doesn't current support the L2 normalization feature yet. L2 normalization is a feature that is currently in the works.

 

However a possible work around in the mean time could be to perform the L2 normalization calculation on the CPU of the host system. This can be done by specifying the output layer as the previous fully connected layer (similar to Yolo) before the L2 normalization when creating the graph. Afterwards, the L2 normalization can be calculated on the CPU.

0 Kudos
idata
Employee
1,012 Views

@Tome_at_Intel How to specify the the output of previous layer? Shall I remove the end layers from the prototxt?

0 Kudos
idata
Employee
1,012 Views

Got it. Using -in and -on.

0 Kudos
idata
Employee
1,012 Views

@gauravmoon kindly guide how you got it working

 

regards,
0 Kudos
Reply